wan2.2 lightx2v error

[unnikrshnan]

Using seed: 1348343256
Loading Text_Encoder...

FileNotFoundError Traceback (most recent call last)
/tmp/ipython-input-228337594.py in <cell line: 0>()
67
68 # with torch.inference_mode():
---> 69 generate_video(
70 image_path=file_uploaded,
71 LoRA_Strength=LoRA_Strength,

2 frames
/content/ComfyUI/folder_paths.py in get_full_path_or_raise(folder_name, filename)
302 full_path = get_full_path(folder_name, filename)
303 if full_path is None:
--> 304 raise FileNotFoundError(f"Model in folder '{folder_name}' with filename '{filename}' not found.")
305 return full_path
306

FileNotFoundError: Model in folder 'text_encoders' with filename 'umt5_xxl_fp8_e4m3fn_scaled.safetensors' not found.

[runapain]

I too started encountering the same error as unnikrshnan in the video generation cell a few hours ago. After specifying the runtime as the October version, I managed to complete video generation just once, but now it stops with insufficient capacity in the environment setup cell. I hope an expert solution will be presented.

The notebook used was the ▶https://colab.research.google.com/github/Isi-dev/Google-Colab_Notebooks/blob/main/wan2_2/wan22_FirstLastFrame2Video.ipynb.

I am a user from Japan. I am always grateful to the creators of the notebook for teaching me, someone with absolutely no knowledge, how to generate videos for free over the internet. Thank you!

[unnikrshnan]

SO CANT FIX IT ?

[Isi-dev]

Using seed: 1348343256

Loading Text_Encoder...
FileNotFoundError Traceback (most recent call last) /tmp/ipython-input-228337594.py in <cell line: 0>() 67 68 # with torch.inference_mode(): ---> 69 generate_video( 70 image_path=file_uploaded, 71 LoRA_Strength=LoRA_Strength,

2 frames /content/ComfyUI/folder_paths.py in get_full_path_or_raise(folder_name, filename) 302 full_path = get_full_path(folder_name, filename) 303 if full_path is None: --> 304 raise FileNotFoundError(f"Model in folder '{folder_name}' with filename '{filename}' not found.") 305 return full_path 306

FileNotFoundError: Model in folder 'text_encoders' with filename 'umt5_xxl_fp8_e4m3fn_scaled.safetensors' not found.

The FileNotFoundError error indicates that the text encoder was not downloaded. Huggingface model downloads seem to be unstable recently probably due to changes in their download architecture. I have experienced this before and I had to run the 'Prepare Environment' section several times before all the required models were downloaded. I will find out if there's a way to make the model downloads more stable.

[unnikrshnan]

Okay thanq.,,😊

[unnikrshnan]

i have one, can use low noise and high noise lora in your wan2.2 lightx2v?

[runapain]

Was it an issue on Hugging Face's end? Today, I've been able to generate videos smoothly without encountering this error that kept popping up repeatedly. Just wanted to report that.

[unnikrshnan]

yes today working fine.

[Isi-dev]

i have one, can use low noise and high noise lora in your wan2.2 lightx2v?

Yes, you can use low noise and high noise loras in the wan2.2 lightx2v, but the current code applies each LoRA to both models. The code can be modified to apply only high noise loras to the high noise model and only low noise loras to the low noise model.

[Isi-dev]

Was it an issue on Hugging Face's end? Today, I've been able to generate videos smoothly without encountering this error that kept popping up repeatedly. Just wanted to report that.

Thanks for the feedback. It's mostly an issue on HuggingFace's end, but it could also be due to an unstable network.

[unnikrshnan]

i have one, can use low noise and high noise lora in your wan2.2 lightx2v?

Yes, you can use low noise and high noise loras in the wan2.2 lightx2v, but the current code applies each LoRA to both models. The code can be modified to apply only high noise loras to the high noise model and only low noise loras to the low noise model.

okay, now i am use only high noise lora. is it okay? any other tips?

[Isi-dev]

i have one, can use low noise and high noise lora in your wan2.2 lightx2v?

Yes, you can use low noise and high noise loras in the wan2.2 lightx2v, but the current code applies each LoRA to both models. The code can be modified to apply only high noise loras to the high noise model and only low noise loras to the low noise model.

okay, now i am use only high noise lora. is it okay? any other tips?

Yes it is okay as long as the lora is geared towards improving the structure, composition, or motion of people or objects.
Tips: Although using 4 steps for the full generation is a common practice to reduce generation time, you'll get far better results with 6 steps (3 steps each for the high & low noise models).
The 'First-Last Frame to Video with Wan2.2' notebook also does normal image to video generation and it gives you the option of downloading & using other wan2.2 based models and speed loras.

[unnikrshnan]

okay, thanq and appreciate your effort.